Vector database
VEK-tor DAY-tuh-bayss
A database optimized for storing and searching embeddings. The backbone of every RAG pipeline and semantic search system.
A vector database stores embeddings and lets you search them by similarity. Traditional databases search by exact match: find all rows where status equals "active." Vector databases search by closeness: find the ten documents most similar to this query.
The technical term is "nearest neighbor search." You give the database a query vector and it returns the vectors closest to it, measured by cosine similarity or Euclidean distance. This is how RAG retrieval works. This is how semantic search works. Anytime you need to find "things similar to this thing," you need a vector database.
The market is crowded. Pinecone is the dedicated vector database leader. Weaviate and Qdrant are open-source alternatives. Supabase added pgvector to PostgreSQL, so you can store vectors alongside your regular data. For most startups, pgvector in your existing Postgres database is enough. You do not need a separate database until you hit tens of millions of vectors.
Examples
A startup builds a RAG-powered chatbot.
They embed 10,000 documentation pages and store them in Pinecone. When a user asks a question, the system embeds the question, queries Pinecone for the 5 most similar documents, and passes them to Claude for answer generation. Pinecone handles the similarity search in milliseconds.
Using pgvector in an existing Postgres stack.
A Supabase customer adds a vector column to their documents table. They store embeddings alongside titles, content, and metadata in the same database. No new infrastructure. No new vendor. Semantic search is a SQL query with a cosine distance operator.
Scaling vector search.
A company with 50 million product descriptions outgrows pgvector. Query latency climbs to 200ms. They migrate to Pinecone, which uses specialized indexes for billion-scale vector search. Latency drops to 20ms. The tradeoff is another service to manage and pay for.
In practice
Read more on the blog
Related terms
Numerical representations of text that capture semantic meaning. Two similar sentences produce similar numbers, enabling AI-powered search.
Fetching relevant data and feeding it to an LLM so the response is grounded in real, current information instead of training data alone.
A neural network trained on massive text data to generate and understand language. The technology behind ChatGPT, Claude, and Gemini.
Want the complete playbook?
Picks and Shovels is the definitive guide to developer marketing. Amazon #1 bestseller with practical strategies from 30 years of marketing to developers.